SWCompression/Sources/TAR/TarReader.swift

// Copyright (c) 2026 Timofey Solomko
// Licensed under MIT License
//
// See LICENSE for license information

import Foundation
import BitByteData

/**
 A type that allows to iteratively read TAR entries from a container provided by a `FileHandle`.

 The `TarReader` may be helpful in reducing the peak memory usage on certain platforms. However, to achieve this either
 the `TarReader.process(_:)` function should be used or both the call to `TarReader.read()` and the processing of the
 returned entry should be wrapped inside the `autoreleasepool`. Since the `autoreleasepool` is available only on Darwin
 platforms, the memory reducing effect may be not as significant on non-Darwin platforms (such as Linux or Windows).

 The following code demonstrates an example usage of the `TarReader`:
 ```swift
    let handle: FileHandle = ...
    let reader = TarReader(fileHandle: handle)
    try reader.process { ... }
    ...
    try handle.close()
 ```
 Note that closing the `FileHandle` remains the responsibility of the caller.
 */
public struct TarReader {

    private let handle: FileHandle
    private var lastGlobalExtendedHeader: TarExtendedHeader?
    private var lastLocalExtendedHeader: TarExtendedHeader?
    private var longLinkName: String?
    private var longName: String?

    /**
     Creates a new instance for reading TAR entries from the provided `fileHandle`.

     - Parameter fileHandle: A handle from which the entries will be read. Note that the `TarReader` does not close the
     `fileHandle` and this remains the responsibility of the caller.
     */
    public init(fileHandle: FileHandle) {
        self.handle = fileHandle
        self.lastGlobalExtendedHeader = nil
        self.lastLocalExtendedHeader = nil
        self.longLinkName = nil
        self.longName = nil
    }

    /**
     Processes the next TAR entry by reading it from the container and calling the provided closure on the result. If
     the argument supplied to the closure is `nil` this indicates that the end of the input was reached.

     On Darwin platforms both the reading and the call to the closure are performed inside the `autoreleasepool` which
     allows to reduce the peak memory usage.

     - Throws: `DataError.truncated` if the input is truncated. `TarError` is thrown in case of malformed input. Errors
     thrown by `FileHandle` operations are also propagated.
     */
    public mutating func process<T>(_ transform: (TarEntry?) throws -> T) throws -> T {
        return try autoreleasepool {
            let entry = try read()
            return try transform(entry)
        }
    }

    /**
     Reads the next TAR entry from the container.

     On Darwin platforms it is recommended to wrap both the call to this function and the follow-up processing inside
     the `autoreleasepool` in order to reduce the peak memory usage.

     - Throws: `DataError.truncated` if the input is truncated. `TarError` is thrown in case of malformed input. Errors
     thrown by `FileHandle` operations are also propagated.

     - Returns: The next entry from the container or `nil` if the end of the input has been reached.
     */
    public mutating func read() throws -> TarEntry? {
        let headerData = try getData(size: 512)
        if headerData.count == 0 {
            return nil
        } else if headerData == Data(count: 512) {
            // EOF marker case.
            let offset = try handle.offset()
            if try getData(size: 512) == Data(count: 512) {
                return nil
            } else {
                // In this case we have a zero-filled block immediately followed by a non-zero-filled block which do not
                // match the EOF marker signature. In practice, this indicates a malformed TAR container, since a
                // zero-filled block is not a valid TAR header (and in fact the end result is an error being thrown in
                // TarHeader initializer later down the line).
                try handle.seek(toOffset: offset)
            }
        } else if headerData.count < 512 {
            throw DataError.truncated
        }
        assert(headerData.count == 512)

        let header = try TarHeader(LittleEndianByteReader(data: headerData))
        // Since we explicitly initialize the header from 512 bytes-long Data, we don't have to check that we processed
        // at most 512 bytes.
        // Check, just in case, since we use blockStartIndex = -1 when creating TAR containers.
        assert(header.blockStartIndex >= 0)
        let dataStartOffset = try handle.offset()

        let entryData = try getData(size: header.size)
        guard entryData.count == header.size
            else { throw DataError.truncated }

        if case .special(let specialEntryType) = header.type {
            switch specialEntryType {
            case .globalExtendedHeader:
                lastGlobalExtendedHeader = try TarExtendedHeader(entryData)
            case .sunExtendedHeader:
                fallthrough
            case .localExtendedHeader:
                lastLocalExtendedHeader = try TarExtendedHeader(entryData)
            case .longLinkName:
                longLinkName = LittleEndianByteReader(data: entryData).tarCString(maxLength: header.size)
            case .longName:
                longName = LittleEndianByteReader(data: entryData).tarCString(maxLength: header.size)
            }
            try handle.seek(toOffset: dataStartOffset + UInt64(truncatingIfNeeded: header.size.roundTo512()))
            return try read()
        } else {
            let info = TarEntryInfo(header, lastGlobalExtendedHeader, lastLocalExtendedHeader, longName, longLinkName)
            try handle.seek(toOffset: dataStartOffset + UInt64(truncatingIfNeeded: header.size.roundTo512()))
            lastLocalExtendedHeader = nil
            longName = nil
            longLinkName = nil
            if info.type == .directory {
                // For directories TarEntry.data is set to nil.
                var entry = TarEntry(info: info, data: nil)
                entry.info.size = 0
                return entry
            } else {
                return TarEntry(info: info, data: entryData)
            }
        }
    }

    @inline(__always)
    private func getData(size: Int) throws -> Data {
        assert(size >= 0, "TarReader.getData(size:): negative size.")
        // The documentation for FileHandle.read(upToCount:) is a bit misleading. This method does "return the data
        // obtained by reading length bytes starting at the current file pointer" even if the requested amount is
        // larger than the available data. What is not clear is when the method returns nil. Apparently, there are
        // (at least) two cases when it happens:
        //  - the file pointer is at the EOF regardless of the argument value,
        //  - the argument is zero.
        // It is also unclear what happens when the argument is negative (it seems that it reads everything until
        // the EOF), but the assertion above takes care of this. In any case, instead of returning nil we return
        // empty data since both of these situations logically seem equivalent for our purposes. This also allows us
        // to eliminate additional guard-check for the size parameter.
        return try handle.read(upToCount: size) ?? Data()
    }

}

#if os(Linux) || os(Windows) || os(Android) || os(WASI)
@discardableResult
fileprivate func autoreleasepool<T>(_ block: () throws -> T) rethrows -> T {
    return try block()
}
#endif